struct extent_buffer **must_clean);
static int find_next_key(struct btrfs_path *path, int level,
struct btrfs_key *key);
+static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
+ int dump_block_groups);
static noinline int
block_group_cache_done(struct btrfs_block_group_cache *cache)
int stripe_len;
int i, nr, ret;
+ if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
+ stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
+ cache->bytes_super += stripe_len;
+ ret = add_excluded_extent(root, cache->key.objectid,
+ stripe_len);
+ BUG_ON(ret);
+ }
+
for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
bytenr = btrfs_sb_offset(i);
ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
if (ret)
break;
- if (extent_start == start) {
+ if (extent_start <= start) {
start = extent_end + 1;
} else if (extent_start > start && extent_start < end) {
size = extent_start - start;
return ret;
}
-#ifdef BIO_RW_DISCARD
static void btrfs_issue_discard(struct block_device *bdev,
u64 start, u64 len)
{
blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
DISCARD_FL_BARRIER);
}
-#endif
static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
u64 num_bytes)
{
-#ifdef BIO_RW_DISCARD
int ret;
u64 map_length = num_bytes;
struct btrfs_multi_bio *multi = NULL;
+ if (!btrfs_test_opt(root, DISCARD))
+ return 0;
+
/* Tell the block device(s) that the sectors can be discarded */
ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
bytenr, &map_length, &multi, 0);
}
return ret;
-#else
- return 0;
-#endif
}
int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
alloc_target);
}
+static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items)
+{
+ u64 num_bytes;
+ int level;
+
+ level = BTRFS_MAX_LEVEL - 2;
+ /*
+ * NOTE: these calculations are absolutely the worst possible case.
+ * This assumes that _every_ item we insert will require a new leaf, and
+ * that the tree has grown to its maximum level size.
+ */
+
+ /*
+ * for every item we insert we could insert both an extent item and a
+ * extent ref item. Then for ever item we insert, we will need to cow
+ * both the original leaf, plus the leaf to the left and right of it.
+ *
+ * Unless we are talking about the extent root, then we just want the
+ * number of items * 2, since we just need the extent item plus its ref.
+ */
+ if (root == root->fs_info->extent_root)
+ num_bytes = num_items * 2;
+ else
+ num_bytes = (num_items + (2 * num_items)) * 3;
+
+ /*
+ * num_bytes is total number of leaves we could need times the leaf
+ * size, and then for every leaf we could end up cow'ing 2 nodes per
+ * level, down to the leaf level.
+ */
+ num_bytes = (num_bytes * root->leafsize) +
+ (num_bytes * (level * 2)) * root->nodesize;
+
+ return num_bytes;
+}
+
/*
- * for now this just makes sure we have at least 5% of our metadata space free
- * for use.
+ * Unreserve metadata space for delalloc. If we have less reserved credits than
+ * we have extents, this function does nothing.
*/
-int btrfs_check_metadata_free_space(struct btrfs_root *root)
+int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
+ struct inode *inode, int num_items)
{
struct btrfs_fs_info *info = root->fs_info;
struct btrfs_space_info *meta_sinfo;
- u64 alloc_target, thresh;
- int committed = 0, ret;
+ u64 num_bytes;
+ u64 alloc_target;
+ bool bug = false;
/* get the space info for where the metadata will live */
alloc_target = btrfs_get_alloc_profile(root, 0);
meta_sinfo = __find_space_info(info, alloc_target);
- if (!meta_sinfo)
- goto alloc;
-again:
+ num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
+ num_items);
+
spin_lock(&meta_sinfo->lock);
- if (!meta_sinfo->full)
- thresh = meta_sinfo->total_bytes * 80;
- else
- thresh = meta_sinfo->total_bytes * 95;
+ spin_lock(&BTRFS_I(inode)->accounting_lock);
+ if (BTRFS_I(inode)->reserved_extents <=
+ BTRFS_I(inode)->outstanding_extents) {
+ spin_unlock(&BTRFS_I(inode)->accounting_lock);
+ spin_unlock(&meta_sinfo->lock);
+ return 0;
+ }
+ spin_unlock(&BTRFS_I(inode)->accounting_lock);
+
+ BTRFS_I(inode)->reserved_extents--;
+ BUG_ON(BTRFS_I(inode)->reserved_extents < 0);
+
+ if (meta_sinfo->bytes_delalloc < num_bytes) {
+ bug = true;
+ meta_sinfo->bytes_delalloc = 0;
+ } else {
+ meta_sinfo->bytes_delalloc -= num_bytes;
+ }
+ spin_unlock(&meta_sinfo->lock);
+
+ BUG_ON(bug);
+ return 0;
+}
+
+static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
+{
+ u64 thresh;
+
+ thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
+ meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
+ meta_sinfo->bytes_super + meta_sinfo->bytes_root +
+ meta_sinfo->bytes_may_use;
+
+ thresh = meta_sinfo->total_bytes - thresh;
+ thresh *= 80;
do_div(thresh, 100);
+ if (thresh <= meta_sinfo->bytes_delalloc)
+ meta_sinfo->force_delalloc = 1;
+ else
+ meta_sinfo->force_delalloc = 0;
+}
- if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
- meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
- meta_sinfo->bytes_super > thresh) {
- struct btrfs_trans_handle *trans;
- if (!meta_sinfo->full) {
- meta_sinfo->force_alloc = 1;
+struct async_flush {
+ struct btrfs_root *root;
+ struct btrfs_space_info *info;
+ struct btrfs_work work;
+};
+
+static noinline void flush_delalloc_async(struct btrfs_work *work)
+{
+ struct async_flush *async;
+ struct btrfs_root *root;
+ struct btrfs_space_info *info;
+
+ async = container_of(work, struct async_flush, work);
+ root = async->root;
+ info = async->info;
+
+ btrfs_start_delalloc_inodes(root, 0);
+ wake_up(&info->flush_wait);
+ btrfs_wait_ordered_extents(root, 0, 0);
+
+ spin_lock(&info->lock);
+ info->flushing = 0;
+ spin_unlock(&info->lock);
+ wake_up(&info->flush_wait);
+
+ kfree(async);
+}
+
+static void wait_on_flush(struct btrfs_space_info *info)
+{
+ DEFINE_WAIT(wait);
+ u64 used;
+
+ while (1) {
+ prepare_to_wait(&info->flush_wait, &wait,
+ TASK_UNINTERRUPTIBLE);
+ spin_lock(&info->lock);
+ if (!info->flushing) {
+ spin_unlock(&info->lock);
+ break;
+ }
+
+ used = info->bytes_used + info->bytes_reserved +
+ info->bytes_pinned + info->bytes_readonly +
+ info->bytes_super + info->bytes_root +
+ info->bytes_may_use + info->bytes_delalloc;
+ if (used < info->total_bytes) {
+ spin_unlock(&info->lock);
+ break;
+ }
+ spin_unlock(&info->lock);
+ schedule();
+ }
+ finish_wait(&info->flush_wait, &wait);
+}
+
+static void flush_delalloc(struct btrfs_root *root,
+ struct btrfs_space_info *info)
+{
+ struct async_flush *async;
+ bool wait = false;
+
+ spin_lock(&info->lock);
+
+ if (!info->flushing) {
+ info->flushing = 1;
+ init_waitqueue_head(&info->flush_wait);
+ } else {
+ wait = true;
+ }
+
+ spin_unlock(&info->lock);
+
+ if (wait) {
+ wait_on_flush(info);
+ return;
+ }
+
+ async = kzalloc(sizeof(*async), GFP_NOFS);
+ if (!async)
+ goto flush;
+
+ async->root = root;
+ async->info = info;
+ async->work.func = flush_delalloc_async;
+
+ btrfs_queue_worker(&root->fs_info->enospc_workers,
+ &async->work);
+ wait_on_flush(info);
+ return;
+
+flush:
+ btrfs_start_delalloc_inodes(root, 0);
+ btrfs_wait_ordered_extents(root, 0, 0);
+
+ spin_lock(&info->lock);
+ info->flushing = 0;
+ spin_unlock(&info->lock);
+ wake_up(&info->flush_wait);
+}
+
+static int maybe_allocate_chunk(struct btrfs_root *root,
+ struct btrfs_space_info *info)
+{
+ struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
+ struct btrfs_trans_handle *trans;
+ bool wait = false;
+ int ret = 0;
+ u64 min_metadata;
+ u64 free_space;
+
+ free_space = btrfs_super_total_bytes(disk_super);
+ /*
+ * we allow the metadata to grow to a max of either 10gb or 5% of the
+ * space in the volume.
+ */
+ min_metadata = min((u64)10 * 1024 * 1024 * 1024,
+ div64_u64(free_space * 5, 100));
+ if (info->total_bytes >= min_metadata) {
+ spin_unlock(&info->lock);
+ return 0;
+ }
+
+ if (info->full) {
+ spin_unlock(&info->lock);
+ return 0;
+ }
+
+ if (!info->allocating_chunk) {
+ info->force_alloc = 1;
+ info->allocating_chunk = 1;
+ init_waitqueue_head(&info->allocate_wait);
+ } else {
+ wait = true;
+ }
+
+ spin_unlock(&info->lock);
+
+ if (wait) {
+ wait_event(info->allocate_wait,
+ !info->allocating_chunk);
+ return 1;
+ }
+
+ trans = btrfs_start_transaction(root, 1);
+ if (!trans) {
+ ret = -ENOMEM;
+ goto out;
+ }
+
+ ret = do_chunk_alloc(trans, root->fs_info->extent_root,
+ 4096 + 2 * 1024 * 1024,
+ info->flags, 0);
+ btrfs_end_transaction(trans, root);
+ if (ret)
+ goto out;
+out:
+ spin_lock(&info->lock);
+ info->allocating_chunk = 0;
+ spin_unlock(&info->lock);
+ wake_up(&info->allocate_wait);
+
+ if (ret)
+ return 0;
+ return 1;
+}
+
+/*
+ * Reserve metadata space for delalloc.
+ */
+int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
+ struct inode *inode, int num_items)
+{
+ struct btrfs_fs_info *info = root->fs_info;
+ struct btrfs_space_info *meta_sinfo;
+ u64 num_bytes;
+ u64 used;
+ u64 alloc_target;
+ int flushed = 0;
+ int force_delalloc;
+
+ /* get the space info for where the metadata will live */
+ alloc_target = btrfs_get_alloc_profile(root, 0);
+ meta_sinfo = __find_space_info(info, alloc_target);
+
+ num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
+ num_items);
+again:
+ spin_lock(&meta_sinfo->lock);
+
+ force_delalloc = meta_sinfo->force_delalloc;
+
+ if (unlikely(!meta_sinfo->bytes_root))
+ meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
+
+ if (!flushed)
+ meta_sinfo->bytes_delalloc += num_bytes;
+
+ used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
+ meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
+ meta_sinfo->bytes_super + meta_sinfo->bytes_root +
+ meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
+
+ if (used > meta_sinfo->total_bytes) {
+ flushed++;
+
+ if (flushed == 1) {
+ if (maybe_allocate_chunk(root, meta_sinfo))
+ goto again;
+ flushed++;
+ } else {
spin_unlock(&meta_sinfo->lock);
-alloc:
- trans = btrfs_start_transaction(root, 1);
- if (!trans)
- return -ENOMEM;
+ }
- ret = do_chunk_alloc(trans, root->fs_info->extent_root,
- 2 * 1024 * 1024, alloc_target, 0);
- btrfs_end_transaction(trans, root);
- if (!meta_sinfo) {
- meta_sinfo = __find_space_info(info,
- alloc_target);
- }
+ if (flushed == 2) {
+ filemap_flush(inode->i_mapping);
+ goto again;
+ } else if (flushed == 3) {
+ flush_delalloc(root, meta_sinfo);
goto again;
}
+ spin_lock(&meta_sinfo->lock);
+ meta_sinfo->bytes_delalloc -= num_bytes;
spin_unlock(&meta_sinfo->lock);
+ printk(KERN_ERR "enospc, has %d, reserved %d\n",
+ BTRFS_I(inode)->outstanding_extents,
+ BTRFS_I(inode)->reserved_extents);
+ dump_space_info(meta_sinfo, 0, 0);
+ return -ENOSPC;
+ }
- if (!committed) {
- committed = 1;
- trans = btrfs_join_transaction(root, 1);
- if (!trans)
- return -ENOMEM;
- ret = btrfs_commit_transaction(trans, root);
- if (ret)
- return ret;
+ BTRFS_I(inode)->reserved_extents++;
+ check_force_delalloc(meta_sinfo);
+ spin_unlock(&meta_sinfo->lock);
+
+ if (!flushed && force_delalloc)
+ filemap_flush(inode->i_mapping);
+
+ return 0;
+}
+
+/*
+ * unreserve num_items number of items worth of metadata space. This needs to
+ * be paired with btrfs_reserve_metadata_space.
+ *
+ * NOTE: if you have the option, run this _AFTER_ you do a
+ * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
+ * oprations which will result in more used metadata, so we want to make sure we
+ * can do that without issue.
+ */
+int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items)
+{
+ struct btrfs_fs_info *info = root->fs_info;
+ struct btrfs_space_info *meta_sinfo;
+ u64 num_bytes;
+ u64 alloc_target;
+ bool bug = false;
+
+ /* get the space info for where the metadata will live */
+ alloc_target = btrfs_get_alloc_profile(root, 0);
+ meta_sinfo = __find_space_info(info, alloc_target);
+
+ num_bytes = calculate_bytes_needed(root, num_items);
+
+ spin_lock(&meta_sinfo->lock);
+ if (meta_sinfo->bytes_may_use < num_bytes) {
+ bug = true;
+ meta_sinfo->bytes_may_use = 0;
+ } else {
+ meta_sinfo->bytes_may_use -= num_bytes;
+ }
+ spin_unlock(&meta_sinfo->lock);
+
+ BUG_ON(bug);
+
+ return 0;
+}
+
+/*
+ * Reserve some metadata space for use. We'll calculate the worste case number
+ * of bytes that would be needed to modify num_items number of items. If we
+ * have space, fantastic, if not, you get -ENOSPC. Please call
+ * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
+ * items you reserved, since whatever metadata you needed should have already
+ * been allocated.
+ *
+ * This will commit the transaction to make more space if we don't have enough
+ * metadata space. THe only time we don't do this is if we're reserving space
+ * inside of a transaction, then we will just return -ENOSPC and it is the
+ * callers responsibility to handle it properly.
+ */
+int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items)
+{
+ struct btrfs_fs_info *info = root->fs_info;
+ struct btrfs_space_info *meta_sinfo;
+ u64 num_bytes;
+ u64 used;
+ u64 alloc_target;
+ int retries = 0;
+
+ /* get the space info for where the metadata will live */
+ alloc_target = btrfs_get_alloc_profile(root, 0);
+ meta_sinfo = __find_space_info(info, alloc_target);
+
+ num_bytes = calculate_bytes_needed(root, num_items);
+again:
+ spin_lock(&meta_sinfo->lock);
+
+ if (unlikely(!meta_sinfo->bytes_root))
+ meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
+
+ if (!retries)
+ meta_sinfo->bytes_may_use += num_bytes;
+
+ used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
+ meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
+ meta_sinfo->bytes_super + meta_sinfo->bytes_root +
+ meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
+
+ if (used > meta_sinfo->total_bytes) {
+ retries++;
+ if (retries == 1) {
+ if (maybe_allocate_chunk(root, meta_sinfo))
+ goto again;
+ retries++;
+ } else {
+ spin_unlock(&meta_sinfo->lock);
+ }
+
+ if (retries == 2) {
+ flush_delalloc(root, meta_sinfo);
goto again;
}
+ spin_lock(&meta_sinfo->lock);
+ meta_sinfo->bytes_may_use -= num_bytes;
+ spin_unlock(&meta_sinfo->lock);
+
+ dump_space_info(meta_sinfo, 0, 0);
return -ENOSPC;
}
+
+ check_force_delalloc(meta_sinfo);
spin_unlock(&meta_sinfo->lock);
return 0;
spin_unlock(&data_sinfo->lock);
/* commit the current transaction and try again */
- if (!committed) {
+ if (!committed && !root->fs_info->open_ioctl_trans) {
committed = 1;
trans = btrfs_join_transaction(root, 1);
if (!trans)
BTRFS_I(inode)->reserved_bytes += bytes;
spin_unlock(&data_sinfo->lock);
- return btrfs_check_metadata_free_space(root);
+ return 0;
}
/*
BUG_ON(!space_info);
spin_lock(&space_info->lock);
- if (space_info->force_alloc) {
+ if (space_info->force_alloc)
force = 1;
- space_info->force_alloc = 0;
- }
if (space_info->full) {
spin_unlock(&space_info->lock);
goto out;
}
thresh = space_info->total_bytes - space_info->bytes_readonly;
- thresh = div_factor(thresh, 6);
+ thresh = div_factor(thresh, 8);
if (!force &&
(space_info->bytes_used + space_info->bytes_pinned +
space_info->bytes_reserved + alloc_bytes) < thresh) {
* we keep a reasonable number of metadata chunks allocated in the
* FS as well.
*/
- if (flags & BTRFS_BLOCK_GROUP_DATA) {
+ if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
fs_info->data_chunk_allocations++;
if (!(fs_info->data_chunk_allocations %
fs_info->metadata_ratio))
}
ret = btrfs_alloc_chunk(trans, extent_root, flags);
+ spin_lock(&space_info->lock);
if (ret)
space_info->full = 1;
+ space_info->force_alloc = 0;
+ spin_unlock(&space_info->lock);
out:
mutex_unlock(&extent_root->fs_info->chunk_mutex);
return ret;
else
old_val -= num_bytes;
btrfs_set_super_bytes_used(&info->super_copy, old_val);
-
- /* block accounting for root item */
- old_val = btrfs_root_used(&root->root_item);
- if (alloc)
- old_val += num_bytes;
- else
- old_val -= num_bytes;
- btrfs_set_root_used(&root->root_item, old_val);
spin_unlock(&info->delalloc_lock);
while (total) {
if (is_data)
goto pinit;
+ /*
+ * discard is sloooow, and so triggering discards on
+ * individual btree blocks isn't a good plan. Just
+ * pin everything in discard mode.
+ */
+ if (btrfs_test_opt(root, DISCARD))
+ goto pinit;
+
buf = btrfs_find_tree_block(root, bytenr, num_bytes);
if (!buf)
goto pinit;
return ret;
}
+int btrfs_free_tree_block(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root,
+ u64 bytenr, u32 blocksize,
+ u64 parent, u64 root_objectid, int level)
+{
+ u64 used;
+ spin_lock(&root->node_lock);
+ used = btrfs_root_used(&root->root_item) - blocksize;
+ btrfs_set_root_used(&root->root_item, used);
+ spin_unlock(&root->node_lock);
+
+ return btrfs_free_extent(trans, root, bytenr, blocksize,
+ parent, root_objectid, level, 0);
+}
+
static u64 stripe_align(struct btrfs_root *root, u64 val)
{
u64 mask = ((u64)root->stripesize - 1);
}
enum btrfs_loop_type {
- LOOP_CACHED_ONLY = 0,
+ LOOP_FIND_IDEAL = 0,
LOOP_CACHING_NOWAIT = 1,
LOOP_CACHING_WAIT = 2,
LOOP_ALLOC_CHUNK = 3,
struct btrfs_block_group_cache *block_group = NULL;
int empty_cluster = 2 * 1024 * 1024;
int allowed_chunk_alloc = 0;
+ int done_chunk_alloc = 0;
struct btrfs_space_info *space_info;
int last_ptr_loop = 0;
int loop = 0;
bool found_uncached_bg = false;
bool failed_cluster_refill = false;
+ bool failed_alloc = false;
+ u64 ideal_cache_percent = 0;
+ u64 ideal_cache_offset = 0;
WARN_ON(num_bytes < root->sectorsize);
btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
empty_cluster = 0;
if (search_start == hint_byte) {
+ideal_cache:
block_group = btrfs_lookup_block_group(root->fs_info,
search_start);
/*
* we don't want to use the block group if it doesn't match our
* allocation bits, or if its not cached.
+ *
+ * However if we are re-searching with an ideal block group
+ * picked out then we don't care that the block group is cached.
*/
if (block_group && block_group_bits(block_group, data) &&
- block_group_cache_done(block_group)) {
+ (block_group->cached != BTRFS_CACHE_NO ||
+ search_start == ideal_cache_offset)) {
down_read(&space_info->groups_sem);
if (list_empty(&block_group->list) ||
block_group->ro) {
*/
btrfs_put_block_group(block_group);
up_read(&space_info->groups_sem);
- } else
+ } else {
goto have_block_group;
+ }
} else if (block_group) {
btrfs_put_block_group(block_group);
}
}
-
search:
down_read(&space_info->groups_sem);
list_for_each_entry(block_group, &space_info->block_groups, list) {
have_block_group:
if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
+ u64 free_percent;
+
+ free_percent = btrfs_block_group_used(&block_group->item);
+ free_percent *= 100;
+ free_percent = div64_u64(free_percent,
+ block_group->key.offset);
+ free_percent = 100 - free_percent;
+ if (free_percent > ideal_cache_percent &&
+ likely(!block_group->ro)) {
+ ideal_cache_offset = block_group->key.objectid;
+ ideal_cache_percent = free_percent;
+ }
+
/*
- * we want to start caching kthreads, but not too many
- * right off the bat so we don't overwhelm the system,
- * so only start them if there are less than 2 and we're
- * in the initial allocation phase.
+ * We only want to start kthread caching if we are at
+ * the point where we will wait for caching to make
+ * progress, or if our ideal search is over and we've
+ * found somebody to start caching.
*/
if (loop > LOOP_CACHING_NOWAIT ||
- atomic_read(&space_info->caching_threads) < 2) {
+ (loop > LOOP_FIND_IDEAL &&
+ atomic_read(&space_info->caching_threads) < 2)) {
ret = cache_block_group(block_group);
BUG_ON(ret);
}
- }
-
- cached = block_group_cache_done(block_group);
- if (unlikely(!cached)) {
found_uncached_bg = true;
- /* if we only want cached bgs, loop */
- if (loop == LOOP_CACHED_ONLY)
+ /*
+ * If loop is set for cached only, try the next block
+ * group.
+ */
+ if (loop == LOOP_FIND_IDEAL)
goto loop;
}
+ cached = block_group_cache_done(block_group);
+ if (unlikely(!cached))
+ found_uncached_bg = true;
+
if (unlikely(block_group->ro))
goto loop;
offset = btrfs_find_space_for_alloc(block_group, search_start,
num_bytes, empty_size);
- if (!offset && (cached || (!cached &&
- loop == LOOP_CACHING_NOWAIT))) {
- goto loop;
- } else if (!offset && (!cached &&
- loop > LOOP_CACHING_NOWAIT)) {
+ /*
+ * If we didn't find a chunk, and we haven't failed on this
+ * block group before, and this block group is in the middle of
+ * caching and we are ok with waiting, then go ahead and wait
+ * for progress to be made, and set failed_alloc to true.
+ *
+ * If failed_alloc is true then we've already waited on this
+ * block group once and should move on to the next block group.
+ */
+ if (!offset && !failed_alloc && !cached &&
+ loop > LOOP_CACHING_NOWAIT) {
wait_block_group_cache_progress(block_group,
- num_bytes + empty_size);
+ num_bytes + empty_size);
+ failed_alloc = true;
goto have_block_group;
+ } else if (!offset) {
+ goto loop;
}
checks:
search_start = stripe_align(root, offset);
break;
loop:
failed_cluster_refill = false;
+ failed_alloc = false;
btrfs_put_block_group(block_group);
}
up_read(&space_info->groups_sem);
- /* LOOP_CACHED_ONLY, only search fully cached block groups
- * LOOP_CACHING_NOWAIT, search partially cached block groups, but
- * dont wait foR them to finish caching
+ /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
+ * for them to make caching progress. Also
+ * determine the best possible bg to cache
+ * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
+ * caching kthreads as we move along
* LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
* LOOP_ALLOC_CHUNK, force a chunk allocation and try again
* LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
(found_uncached_bg || empty_size || empty_cluster ||
allowed_chunk_alloc)) {
- if (found_uncached_bg) {
+ if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
found_uncached_bg = false;
- if (loop < LOOP_CACHING_WAIT) {
- loop++;
+ loop++;
+ if (!ideal_cache_percent &&
+ atomic_read(&space_info->caching_threads))
goto search;
- }
+
+ /*
+ * 1 of the following 2 things have happened so far
+ *
+ * 1) We found an ideal block group for caching that
+ * is mostly full and will cache quickly, so we might
+ * as well wait for it.
+ *
+ * 2) We searched for cached only and we didn't find
+ * anything, and we didn't start any caching kthreads
+ * either, so chances are we will loop through and
+ * start a couple caching kthreads, and then come back
+ * around and just wait for them. This will be slower
+ * because we will have 2 caching kthreads reading at
+ * the same time when we could have just started one
+ * and waited for it to get far enough to give us an
+ * allocation, so go ahead and go to the wait caching
+ * loop.
+ */
+ loop = LOOP_CACHING_WAIT;
+ search_start = ideal_cache_offset;
+ ideal_cache_percent = 0;
+ goto ideal_cache;
+ } else if (loop == LOOP_FIND_IDEAL) {
+ /*
+ * Didn't find a uncached bg, wait on anything we find
+ * next.
+ */
+ loop = LOOP_CACHING_WAIT;
+ goto search;
+ }
+
+ if (loop < LOOP_CACHING_WAIT) {
+ loop++;
+ goto search;
}
if (loop == LOOP_ALLOC_CHUNK) {
ret = do_chunk_alloc(trans, root, num_bytes +
2 * 1024 * 1024, data, 1);
allowed_chunk_alloc = 0;
- } else {
+ done_chunk_alloc = 1;
+ } else if (!done_chunk_alloc) {
space_info->force_alloc = 1;
}
return ret;
}
-static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
+static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
+ int dump_block_groups)
{
struct btrfs_block_group_cache *cache;
+ spin_lock(&info->lock);
printk(KERN_INFO "space_info has %llu free, is %sfull\n",
(unsigned long long)(info->total_bytes - info->bytes_used -
- info->bytes_pinned - info->bytes_reserved),
+ info->bytes_pinned - info->bytes_reserved -
+ info->bytes_super),
(info->full) ? "" : "not ");
printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
- " may_use=%llu, used=%llu\n",
+ " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
+ "\n",
(unsigned long long)info->total_bytes,
(unsigned long long)info->bytes_pinned,
(unsigned long long)info->bytes_delalloc,
(unsigned long long)info->bytes_may_use,
- (unsigned long long)info->bytes_used);
+ (unsigned long long)info->bytes_used,
+ (unsigned long long)info->bytes_root,
+ (unsigned long long)info->bytes_super,
+ (unsigned long long)info->bytes_reserved);
+ spin_unlock(&info->lock);
+
+ if (!dump_block_groups)
+ return;
down_read(&info->groups_sem);
list_for_each_entry(cache, &info->block_groups, list) {
{
int ret;
u64 search_start = 0;
- struct btrfs_fs_info *info = root->fs_info;
data = btrfs_get_alloc_profile(root, data);
again:
* the only place that sets empty_size is btrfs_realloc_node, which
* is not called recursively on allocations
*/
- if (empty_size || root->ref_cows) {
- if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
- ret = do_chunk_alloc(trans, root->fs_info->extent_root,
- 2 * 1024 * 1024,
- BTRFS_BLOCK_GROUP_METADATA |
- (info->metadata_alloc_profile &
- info->avail_metadata_alloc_bits), 0);
- }
+ if (empty_size || root->ref_cows)
ret = do_chunk_alloc(trans, root->fs_info->extent_root,
num_bytes + 2 * 1024 * 1024, data, 0);
- }
WARN_ON(num_bytes < root->sectorsize);
ret = find_free_extent(trans, root, num_bytes, empty_size,
printk(KERN_ERR "btrfs allocation failed flags %llu, "
"wanted %llu\n", (unsigned long long)data,
(unsigned long long)num_bytes);
- dump_space_info(sinfo, num_bytes);
+ dump_space_info(sinfo, num_bytes, 1);
}
return ret;
extent_op);
BUG_ON(ret);
}
+
+ if (root_objectid == root->root_key.objectid) {
+ u64 used;
+ spin_lock(&root->node_lock);
+ used = btrfs_root_used(&root->root_item) + num_bytes;
+ btrfs_set_root_used(&root->root_item, used);
+ spin_unlock(&root->node_lock);
+ }
return ret;
}
btrfs_set_buffer_uptodate(buf);
if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
- set_extent_dirty(&root->dirty_log_pages, buf->start,
- buf->start + buf->len - 1, GFP_NOFS);
+ /*
+ * we allow two log transactions at a time, use different
+ * EXENT bit to differentiate dirty pages.
+ */
+ if (root->log_transid % 2 == 0)
+ set_extent_dirty(&root->dirty_log_pages, buf->start,
+ buf->start + buf->len - 1, GFP_NOFS);
+ else
+ set_extent_new(&root->dirty_log_pages, buf->start,
+ buf->start + buf->len - 1, GFP_NOFS);
} else {
set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
buf->start + buf->len - 1, GFP_NOFS);
u64 bytenr;
u64 generation;
u64 refs;
+ u64 flags;
u64 last = 0;
u32 nritems;
u32 blocksize;
generation <= root->root_key.offset)
continue;
+ /* We don't lock the tree block, it's OK to be racy here */
+ ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
+ &refs, &flags);
+ BUG_ON(ret);
+ BUG_ON(refs == 0);
+
if (wc->stage == DROP_REFERENCE) {
- ret = btrfs_lookup_extent_info(trans, root,
- bytenr, blocksize,
- &refs, NULL);
- BUG_ON(ret);
- BUG_ON(refs == 0);
if (refs == 1)
goto reada;
+ if (wc->level == 1 &&
+ (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
+ continue;
if (!wc->update_ref ||
generation <= root->root_key.offset)
continue;
&wc->update_progress);
if (ret < 0)
continue;
+ } else {
+ if (wc->level == 1 &&
+ (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
+ continue;
}
reada:
ret = readahead_tree_block(root, bytenr, blocksize,
static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
struct btrfs_root *root,
struct btrfs_path *path,
- struct walk_control *wc)
+ struct walk_control *wc, int lookup_info)
{
int level = wc->level;
struct extent_buffer *eb = path->nodes[level];
* when reference count of tree block is 1, it won't increase
* again. once full backref flag is set, we never clear it.
*/
- if ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
- (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag))) {
+ if (lookup_info &&
+ ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
+ (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
BUG_ON(!path->locks[level]);
ret = btrfs_lookup_extent_info(trans, root,
eb->start, eb->len,
static noinline int do_walk_down(struct btrfs_trans_handle *trans,
struct btrfs_root *root,
struct btrfs_path *path,
- struct walk_control *wc)
+ struct walk_control *wc, int *lookup_info)
{
u64 bytenr;
u64 generation;
* for the subtree
*/
if (wc->stage == UPDATE_BACKREF &&
- generation <= root->root_key.offset)
+ generation <= root->root_key.offset) {
+ *lookup_info = 1;
return 1;
+ }
bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
blocksize = btrfs_level_size(root, level - 1);
btrfs_tree_lock(next);
btrfs_set_lock_blocking(next);
- if (wc->stage == DROP_REFERENCE) {
- ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
- &wc->refs[level - 1],
- &wc->flags[level - 1]);
- BUG_ON(ret);
- BUG_ON(wc->refs[level - 1] == 0);
+ ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
+ &wc->refs[level - 1],
+ &wc->flags[level - 1]);
+ BUG_ON(ret);
+ BUG_ON(wc->refs[level - 1] == 0);
+ *lookup_info = 0;
+ if (wc->stage == DROP_REFERENCE) {
if (wc->refs[level - 1] > 1) {
+ if (level == 1 &&
+ (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
+ goto skip;
+
if (!wc->update_ref ||
generation <= root->root_key.offset)
goto skip;
wc->stage = UPDATE_BACKREF;
wc->shared_level = level - 1;
}
+ } else {
+ if (level == 1 &&
+ (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
+ goto skip;
}
if (!btrfs_buffer_uptodate(next, generation)) {
btrfs_tree_unlock(next);
free_extent_buffer(next);
next = NULL;
+ *lookup_info = 1;
}
if (!next) {
skip:
wc->refs[level - 1] = 0;
wc->flags[level - 1] = 0;
+ if (wc->stage == DROP_REFERENCE) {
+ if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
+ parent = path->nodes[level]->start;
+ } else {
+ BUG_ON(root->root_key.objectid !=
+ btrfs_header_owner(path->nodes[level]));
+ parent = 0;
+ }
- if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
- parent = path->nodes[level]->start;
- } else {
- BUG_ON(root->root_key.objectid !=
- btrfs_header_owner(path->nodes[level]));
- parent = 0;
+ ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
+ root->root_key.objectid, level - 1, 0);
+ BUG_ON(ret);
}
-
- ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
- root->root_key.objectid, level - 1, 0);
- BUG_ON(ret);
-
btrfs_tree_unlock(next);
free_extent_buffer(next);
+ *lookup_info = 1;
return 1;
}
struct walk_control *wc)
{
int level = wc->level;
+ int lookup_info = 1;
int ret;
while (level >= 0) {
btrfs_header_nritems(path->nodes[level]))
break;
- ret = walk_down_proc(trans, root, path, wc);
+ ret = walk_down_proc(trans, root, path, wc, lookup_info);
if (ret > 0)
break;
if (level == 0)
break;
- ret = do_walk_down(trans, root, path, wc);
+ ret = do_walk_down(trans, root, path, wc, &lookup_info);
if (ret > 0) {
path->slots[level]++;
continue;
git://ftp.safe.ca / safe / jmp / linux-2.6 / blobdiff